Brief Details: LongVU_Qwen2_7B is a 7.67B parameter video-language understanding model with strong performance on video comprehension tasks, achieving 67.6% accuracy on EgoSchema.
Brief Details: A specialized LoRA model trained on yarn art style images, built on FLUX.1-dev base model. Popular with 14k+ downloads, enables yarn-texture artistic effects.
Brief Details: VFusion3D is a 452M parameter image-to-3D generative model that converts single images into 3D content using video diffusion techniques and transformer architecture.
Brief Details: Quality-Aware Diffusion model for text-to-music generation, featuring masked diffusion transformer (MDT) architecture. State-of-the-art results on MusicCaps.
Brief Details: A specialized 8B parameter Chinese-English instruction-tuned LLaMA model, optimized for GGUF format with multiple quantization options and strong chat capabilities.
Brief Details: DeepSeek-Coder-V2-Lite-Base: 15.7B-parameter code-specialized model with 128k context length, optimized for programming tasks across 338 languages.
Brief Details: A 7B parameter bilingual LLM optimized for Traditional Chinese & English, built on Mistral-7B with expanded vocabulary and strong performance in reasoning and knowledge tasks.
Brief Details: DanTagGen-beta is a 400M parameter LLaMA-based model for generating Danbooru-style image tags, trained on 5.3M datasets with enhanced capabilities for detailed art descriptions.
Brief Details: An advanced image tagging model with 98M parameters, based on SwinV2 architecture. Specializes in anime/manga content tagging with high F1 score (0.4541).
Brief Details: MusicLang-v2 is a controllable symbolic music generation model based on LLAMA2 architecture, enabling MIDI export and chord progression control.
Brief Details: Bilingual text-to-image diffusion model with 3.5B parameters, optimized for both Chinese and English prompts. Features enhanced CLIP-based architecture and superior generation quality.
Brief-details: High-performing 7B parameter Mistral-based model with strong MT Bench (8.51) and EQ Bench (42.18) scores, optimized for text generation in GGUF format.
Brief-details: A specialized 34B parameter financial analysis model trained on high-quality research reports and financial data, focused on quantitative market analysis and investment evaluation.
Brief-details: Advanced 7B parameter LLM based on Mistral, fine-tuned with DPO, optimized for coding tasks and general instruction-following. Shows strong performance across multiple benchmarks with 67.20% average score.
Brief-details: 7B parameter Mistral-based model optimized for roleplay and general tasks, featuring strong MT-Bench scores (7.95) and MMLU performance (~64.9), using DARE TIES merger methodology.
BRIEF DETAILS: Compact 1.1B parameter LLaMA-compatible model trained on 1.5T tokens, offering efficient performance with small footprint. Intermediate checkpoint showing promising evaluation metrics.
Brief Details: A 13B parameter uncensored LLaMA2-based model using the WizardLM architecture, optimized for GGML format with multiple quantization options
Brief Details: Mixtral-8x7B-Instruct v0.1 4-bit quantized model - 24.2B params, optimized for efficient inference using bitsandbytes. Supports text generation and conversation.
BRIEF-DETAILS: A 33B parameter uncensored LLM based on WizardLM, achieving 59.99% average on key benchmarks with strong performance in HellaSwag (83.84%) and Winogrande (77.66%).
Brief-details: Orca Mini 3B GGML is a lightweight, CPU-optimized version of the Orca architecture, trained on explain-tuned datasets with multiple quantization options for efficient deployment.
Brief Details: A 40B parameter uncensored Falcon model quantized to 4-bit precision using GPTQ. Features WizardLM training without alignment constraints, optimized for GPU inference.
